MD Simulations and Experimental Study of Structure, Dynamics, and

Jun 24, 2003 - Oleg Borodin,*,† Richard Douglas,† Grant D. Smith,†,‡ Frans Trouw,§ and Sergio Petrucci#. Department of Materials Science and ...
0 downloads 0 Views 243KB Size
J. Phys. Chem. B 2003, 107, 6813-6823

6813

MD Simulations and Experimental Study of Structure, Dynamics, and Thermodynamics of Poly(ethylene oxide) and Its Oligomers Oleg Borodin,*,† Richard Douglas,† Grant D. Smith,†,‡ Frans Trouw,§ and Sergio Petrucci# Department of Materials Science and Engineering, 122 S. Central Campus DriVe, Rm. 304, UniVersity of Utah, Salt Lake City, Utah 84112, Manuel Lujan Jr. Neutron Scattering Center, MS H805, Los Alamos National Laboratory, New Mexico 87545, Department of Chemical and Fuels Engineering, UniVersity of Utah, Salt Lake City, Utah 84112, and Polytechnic UniVersity, Farmingdale, New York 11735 ReceiVed: NoVember 22, 2002; In Final Form: April 14, 2003

Molecular dynamics (MD) simulations have been performed on poly(ethylene oxide) (PEO) and its oligomers, using quantum chemistry-based force fields with and without many-body polarizable interactions. Inclusion of the many-body polarization in the model resulted in increased populations of the tgt and tgg conformations of 1,2-dimethoxyethane and slightly slower dynamics. Increasing the PEO dipole moment also led to increased populations of the tgt and ttg conformers and slower dynamics. Quasi-elastic neutron scattering, dielectric relaxation, and 13C spin-lattice relaxation experiments have been performed on PEO and its oligomers. New 13 C NMR experiments yielded spin-lattice relaxation times that were ∼2-3 times larger than those in the previous experiments, which is in good agreement with the current simulation results. Good agreement between the MD simulation results using many-body polarizable and two-body nonpolarizable potentials was found with experiments for thermodynamic, transport, structural, and dynamic properties.

I. Introduction Poly(ethylene oxide) (PEO) is an ubiquitous polymer that is used in a wide range of technologically important applications, such as polymer electrolytes, protein partitioning, drug delivery, and aqueous biphasic separation, which has been briefly described in ref 1. We are particularly interested in gaining insight into ion transport and developing predictive tools for understanding the structure-property relationship in polymer electrolytes that are comprised of a PEO-based matrix that has been doped with lithium salts. Accurate description of the structural and dynamic properties of PEO melts and PEO/Li+ interactions is important for the accurate modeling of polymer electrolytes. To systematically approach the development of the polymer electrolyte atomistic model, e.g., the PEO/Li salt force field, we initiated a study that has been summarized in a threepaper series1,2 that involves systematic developments of quantum chemistry-based force fields for polymer melts and polymersalt solutions, force field validation, and studies of the structureproperty relationship in polymer electrolytes. This paper is the second in the series. In the first paper,1 we developed the two-body (TB) nonpolarizable and many-body (MB) polarizable quantum chemistrybased force fields for PEO, with the purpose of using this force field for polymer electrolyte modeling.2 The many-body polarizable interactions have been included in the PEO force field, because they are expected to have a significant contribution to the PEO/Li+ complexation energetics and some effect on the structure and dynamics of PEO melts. Investigation of the effect of the many-body polarization on conformational and dynamic * Author to whom correspondence should be addressed. E-mail: e-mail: [email protected]. † Department of Materials Science and Engineering, University of Utah. ‡ Department of Chemical and Fuels Engineering. § Los Alamos National Laboratory. # Polytechnic University.

properties of PEO melts is presented in this paper, whereas the effect of the many-body polarizable interactions on ion transport and the structure of the polymer electrolytes will be discussed in the next paper.2 In this contribution, we also discuss the influence of the choice of partial charges representing the electrostatic potential around PEO on the conformational and dynamic properties of PEO melts and its oligomers. The reason for this investigation is the inability of the partial charge model to represent, with high accuracy (3 Å) but underestimated the electrostatic potential at a point of the expected Li+ complexation to the 1,2-dimethoxyethane (DME) tgt conformer by ∼10 kcal/mol, out of 50 kcal/mol, as shown in Figure 3 of ref 1. The PEO force field with such a description of the electrostatic potential cannot be used for the accurate prediction of DME/Li+ or PEO/ Li+ complexation energetics. Increased weighting of the electrostatic potential at grid points close to the ether atoms during fitting improved the description of the electrostatic potential in the proximity of a molecule, allowing for more-accurate description of the DME/Li+ and PEO/Li+ complexation energies, but at the expense of overestimating the electrostatic potential far from the DME molecule. Thus, various compromises between accurate descriptions in the proximity of a molecule and far from it are possible. In this contribution, we have used two sets of partial charges. The first set of partial charges was obtained by fitting the electrostatic potential from the force field to that from quantum chemistry, using the absolute value of the electrostatic potential to weight the contribution of grid points during fitting (φ-weighting), whereas the second set of partial charges was obtained in a similar manner but with the square of the electrostatic potential used

10.1021/jp0275387 CCC: $25.00 © 2003 American Chemical Society Published on Web 06/24/2003

6814 J. Phys. Chem. B, Vol. 107, No. 28, 2003

Borodin et al.

TABLE 1: Summary of Differences between Quantum Chemistry-Based Potentials Used in This Work force field

weighting function used to weight grid points in the least-squares fitting of partial charges to reproduce electrostatic grid around PEO oligomers

torsional potential were fit to:

FF-1 FF-2 FF-3

absolute value of the electrostatic potential (φ) a square of the electrostatic potential (φ2) a square of the electrostatic potential (φ2)

quantum chemistry energetics quantum chemistry energetics quantum chemistry energetics with barriers lowered by 0.3-0.9 kcal/mol

for weighting grid point contribution to the electrostatic potential during fitting (φ2-weighting) (see Table 1 and ref 1). The first set of charges used in the FF-1 force fields represents a compromise between descriptions of the electrostatic grip close to and far from PEO oligomers. The quantum chemistry-based FF-1 force field, using this set of partial charges, together with the other quantum chemistry-based parameters, will be shown to give a PEO melt structure and dynamics that are in good agreement with experiments. However, underestimation of the PEO/Li+ complexation energy by the FF-1 force field puts its use for polymer electrolyte modeling in question. The second set of partial charges aims for an accurate description of the PEO/Li+ complexation energy while overestimating the ether dipole moments (by 16%) and electrostatic potential far from the molecule. The FF-2 force field, using this second set of partial charges, will be shown to yield PEO dynamics that are ∼10%-30% slower than the first set of charges, requiring empirical adjustments to the torsional potential to accelerate the polymer dynamics and improve the agreement with experiments and molecular dynamics (MD) results using the FF-1 force field. The force field with these empirical adjustments will be called the FF-3 force field. Table 1 briefly summarizes the differences between the force fields. In this contribution, we report MD simulations of PEO and its oligomers, using these force fields (FF-1, FF-2, and FF-3 MB polarizable force fields and TB nonpolarizable force fields) to (a) investigate the effect of many-body polarizable interaction on the PEO conformations, structure, and dynamics; (b) explore the influence of the choice of partial charges on the PEO conformations, structure, and dynamics; and (c) validate the quantum chemistry-based force fields against experimental data. New quasi-elastic neutron scattering (QNS), dielectric spectroscopy, and 13C spin-lattice relaxation NMR experiments have been performed on PEO and its oligomers to validate the force fields and resolve inconsistencies in some previously published experimental data with MD simulations results and the other experiments. II. Previous MD Simulations of PEO Several PEO force fields have been employed in MD simulations of PEO, its oligomers, and PEO-based polymer electrolytes.3-8 None of these models have included many-body polarizable interactions. Previous MD simulations by Smith et al.5 with a quantum chemistry-based force field yielded the correct temperature dependence of the PEO characteristic ratio and the characteristic ratio itself (C∞ ) 5.5 ( 0.3),9,10 as well as structure factors,11 the position of the maximum of the dielectric loss spectra,12 and spin-lattice relaxation times12 (T1) that are in good agreement with the experiment. However, the 13C spin-lattice relaxation times from NMR experiments reported in this manuscript differ considerably from the previously reported values,12 indicating that the PEO dynamics predicted by that force field are somewhat slow. This issue will be addressed in this work. The Neyertz et al.6,13 force field was validated against the available crystal PEO data. They found a PEO characteristic

ratio of 5.5, which was in good agreement with experiments,10 but their abundant populations of g+g-g+ is inconsistent with our quantum chemistry results, which casts doubt on the accuracy of their force field, as discussed in ref 9. Halley et al.3 compared the structure factor for the PEO from MD simulations with that from the experiment, achieving only limited success. The de Leeuw group used a modified version of the Neyertz et al.6 force field in their MD simulations of PEO oligomers and found good agreement for the intermediate incoherent structure factor between MD simulations and neutron spin-echo (NSE) experiments.14 Other groups4,6,8 limited their validation of their PEO simulations to comparing the densities from MD simulations and experiments, making it difficult to judge the ability of the force fields used to reproduce polymer structural and dynamical properties. III. Experimental Section A. Quasi-elastic Neutron Scattering Measurements on PEO. Poly(ethylene glycol) dimethyl ether, or “PEO-500”, was used for the neutron scattering experiments at a temperature of 318 K. The polymer used was relatively polydisperse (Mw ) 470, Mn ) 398), as obtained from Aldrich, and was used as received. The QNS experiments were performed on the disk chopper spectrometer (DCS) that is located at the NIST Center for Neutron Research (http://www.ncnr.nist.gov). This direct geometry chopper spectrometer was operated with an incident neutron wavelength of 0.48 nm, which yielded an energy resolution function that is approximately Gaussian in shape, with a full width at half-maximum of 105 µeV. An annular sample geometry was chosen for this experiment, in an attempt to reduce the multiple scattering in a thick sample. An aluminum sample can that was composed of two concentric cans was used, with an approximate diameter of 1.98 cm and an annular gap of 0.04 cm. The sample can was mounted to a closed-circuit refrigerator that was equipped with a heater to allow for sample control at 318 K. A more detailed description of the measurements and data analysis is reported elsewhere.15 Sample loading was performed under an argon atmosphere. B. Quasi-elastic Neutron Scattering Measurements on 1,2Dimethoxyethane. These experiments were conducted at the Intense Pulse Neutron Source Division of the Argonne National Laboratory, using the QNS spectrometer,16 which is a “crystalanalyzer” or “inverse-geometry” spectrometer that accepts a white beam from a solid methane moderator onto the sample. Constant-energy focusing of the graphite analyzer crystal arrays provides for an energy resolution of ∼100 µeV for the quasielastic and low-energy inelastic scattering. The complete scan using the rotating table yields nine spectra, covering the momentum transfer range from 0.5 to 2.5 Å-1. 1,2-Dimethoxyethane (DME) was purchased from Aldrich and used without further purification. The samples were evenly distributed in a cylindrical aluminum envelope, to minimize sample thickness and, thus, reduce multiple scattering, and were placed in a circular thin aluminum cylinder ∼12 cm long and 0.7 cm in diameter. An indium gasket seal was used to keep the sample in the container, because the sample is placed in a vacuum

Structure, Dynamics, and Thermodynamics of PEO

J. Phys. Chem. B, Vol. 107, No. 28, 2003 6815

Figure 1. Structure of 13C NMR spectrum of neat PEO.

during the experiment. Experiments were performed at 318 K. Typical data collection time was 6-8 h. The time-of-flight data were then transformed via the neutron scattering law, S(Q,ω), using the standard conversion programs, as described in ref 17. A fast Fourier transform was used to convert S(Q,ω) to an intermediate scattering function, I(Q,t).17 C. 13C Spin-Lattice NMR Measurements. Measurement of the 13C-proton-decoupled NMR at a resonance frequency of 500 MHz was conducted on PEO-500 (Mw ) 470, Mn ) 398; Lot No. 02114CO) and another poly(ethylene glycol) dimethyl ether (“PEO-2000”, Mw ) 1854, Mn ) 1733; Lot No. 350765/1), each of which were obtained from Aldrich and used without further purification. PEO readily absorbs water vapor, oxygen gas, and carbon dioxide gas from the atmosphere, causing unwanted changes that would have an unfavorable presence in experimental results; therefore, all PEO samples were subjected to several freeze-thaw cycles, under vacuum inside the transfer chamber of a dry glovebox operated in an atmosphere of nitrogen gas. After the sample tubes were prepared, they were capped with plastic lids and then flame-sealed with a propaneoxygen torch, to avoid any future contamination. The 13C NMR experiments utilized an Inova 500 MHz spectrometer with a 5-mm pulsed-field-gradient (PFG) switchable broadband probe. Pyrex glass NMR sample tubes (5.0 mm-528pp, 7 in. long) from the Wilmad/Lab Glass Company were used. T1 values were measured by the inversion recovery method, with the 180°-perturbation pulse width calibrated to produce a zero-intensity resonant peak spectrum. Each spectrum obtained was the average of four transients, and a line broadening of 2 Hz was used in the processing of the free induction decay. At the time of the experiment, the variabletemperature controller of the NMR equipment was not calibrated. However, later tests using a ethylene glycol standard demonstrated that the temperature controller was accurate to (3 °C. At 500 MHz, the 13C-proton decoupled spectrum revealed four distinguishable resonance peaks, as shown in Figure 1. The most prominent peak was due to the resonance signals of multiple interior carbon atoms only, excluding the three carbon atoms on both ends. The remaining three, less-intense resonance

peaks are each caused by the carbon atoms at PEO chain ends and the end methyl carbon groups (namely, the methyl-, R-, and β-carbon atoms, working inward from the ends, respectively). The chemical-shift anisotropy observed along the ends of the PEO chains is greatest for the methyl end groups, which also exhibit the longest relaxation times. This trend continues inward to the R- and β-carbon atoms, as well as all other interior carbon atoms, becoming less significant the farther inward the carbon atom is, relative to the end methyl carbon group. The relaxation time T1 for any resonance peak in the spectrum is characterized as the time required for the peak to return to thermal equilibrium after the fully intense peak is completely inverted by a 180° perturbation pulse. Varian software that was applied in conjunction with the NMR device was used to determine the T1 relaxation times via the inversion recovery method. This method employs a graphical representation of an array of different spectra, each having different relative peak intestines, or peak heights, as a function of delay time. The delay time is the time allocated for relaxation before acquisition of the spectrum. The software then performs a nonlinear regression to converge, within a 95% confidence interval, on a value for T1 as an inverse decay constant into a standard exponential spin-lattice decay function, which best fits the curve to match the peak intensity decay pattern for the spectral array. The errors in the reported T1 values are the standard deviation from actual peak intensities to those obtained by the nonlinear regression from the empirically fit decay function (also done using the Varian NMR software). D. Dielectric Spectroscopy Experiments. PEO-500 (Mw ) 470, Mn ) 398) was obtained from Fluka and dried over a molecular sieve. The equipment and procedure for the dielectric measurements have been described previously.18,19 All the sample manipulations were performed in a glovebox that was filled with dry nitrogen. E. Viscosity Measurements. Viscosity measurements were performed at 318-363 K on PEO-500 (Mw ) 470, Mn ) 398) that had been obtained from Aldrich and used without further purification. Two independent sets of measurements were performed, using a Brookfield rotating viscometer and a Cannon-Fenske type, slanted-tube capillary viscometer that was

6816 J. Phys. Chem. B, Vol. 107, No. 28, 2003

Borodin et al.

calibrated with distilled water and standard oil S6 (No. 84104). Results of the measurements were in agreement, to within 2%. IV. MD Simulations Methodology MD simulations were performed on CH3-capped oligomers of ethylene oxide (H-(CH2-O-CH2)n-H, for n ) 1, 2, 3, ..., 54) and a polydisperse system with Mn ) 398 and Mw ) 475 (referenced in this paper as PEO-500). The linear dimension of the simulation cubic box varied from 23.5 Å to 34 Å, depending on the system. The MD simulations code Lucretius was used.20 All systems were created in the gas phase. The box was shrunk in MD simulations, using a Brownian dynamics algorithm21 over a period of 1-3 ns to yield the experimental densities, with subsequent equilibration in the NPT ensemble for 0.5-5 ns, using the velocity Verlet algorithm with a time step of 1 fs.22 A 1-ns constant pressure run was performed at each temperature, to determine the equilibrium density. Subsequent sampling runs in an NVT ensemble were 1-45 ns in duration. Table 1 summarizes the run lengths for ethylene oxide with n ) 1, 2, and 3 and PEO-500, whereas the NVT runs for n ) 54 were 25 ns at 393 K (the FF-3 force field) and 15 ns for the FF-1 and FF-2 force fields, and 15 ns at 343 K for all force fields. A Nose-Hoover thermostat23 and a barostat24 were used to control the temperature and pressure, whereas bond lengths were constrained using the Shake algorithm.25 The Ewald summation method26 was used for treatment of long-range electrostatic forces between partial charges with partial charges and partial charges with induced dipoles for the many-body polarizable potential, and the particle-mesh Ewald (PME) technique27 was used for the simulations using the two-body nonpolarizable potential. A tapering function28 was used to drive induced dipole-induced dipole interactions to zero at the cutoff of 10 Å. A multiple-time-step reversible reference system propagator algorithm was employed,29 with a time step of 0.75 fs for bonding, bending, and torsional motions, a time step of 1.5 fs for nonbonded interactions within a 6.5 Å sphere, and a time step of 3.0 fs for nonbonded interactions within 6.5-10.0 Å and the reciprocal space part of the Ewald and PME summation. V. Property Calculation from MD Simulations A. Enthalpy of Vaporization. Heats of vaporization (∆H) were calculated using eq 1:

∆H ) Uliq - Uvap + RT Uliq

(1)

Uvap

and are the energies of the liquid and gas phases, where respectively, R is the gas constant, and T is temperature. The gas-phase energy was calculated in MD simulations using Brownian dynamics and no intermolecular interactions. B. Static Structure Factor. The static structure factor was calculated using eq 2:

S(Q) ) 1+

( )∑ 1

〈b〉2

n

R,β

xRbRxβbβ

∫0 (gRβ(r) - 1) rc

( ) sin Qr Qr

C. Intermediate Incoherent Dynamic Structure Factor. For isotropic systems such as liquids, the neutron scattering incoherent intermediate scattering function (ISF), I(Q,t), is given by eq 3:30

Iinc(Q,t) )



sin(∆ri(t)Q) ∆ri(t)Q

(3)

where ∆ri(t) is the displacement of atom i after time t, Q is the magnitude of the momentum transfer vector, and the angled brackets denote an average over all time origins for atoms with a significant incoherent cross section (i.e., hydrogen atoms). D. Frequency-Dependent Dielectric Constant. Linear response theory allows us to obtain the complex dielectric permittivity *(ω) ) ′ + i′′ for the PEO melt, using the relationship31

′ + i′′ ) 1 - iω ∆

∫0∞Φ(t) exp(-iωt) dt

(4)

where the dipole moment autocorrelation function (DACF) is given by

Φ(t) )

〈M(0)‚M(t)〉 〈M(0)‚M(0)〉

(5)

where ∆, the relaxation strength, is equal to r - u. Here, M(t) is the dipole moment of the system at time t, V the volume of the system, kB the Boltzmann constant, and T the temperature. The angled brackets (〈〉) denote an ensemble average. The unrelaxed dielectric constant (u) is the dielectric constant that includes all relaxation processes at frequencies higher than the process of interest, i.e., electronic polarization and relaxation due to vibrations and librations, whereas the relaxed dielectric constant (r) is the value obtained after the relaxation process (i.e., dipole orientational relaxation) is complete. E. Spin-Lattice NMR Relaxation Times (T1). NMR spinlattice relaxation times (T1) probe the rate of decay of the C-H vector, which is related to polymer dynamics. The T1 spinlattice relaxation times for the carbon nuclei associated with the methyl-, R-, and β-carbon atoms, as well as the interior carbon atoms, were measured and labeled according to Figure 1. The experimentally determined T1 values are related to the microscopic motion of the C-H vectors through the relationship32

1 ) K[J(ωH - ωC) + 3J(ωC) + 6J(ωH + ωC)] nT1

(6)

where n is the number of attached protons and ωC and ωH are the Larmor (angular) frequencies of the 13C and 1H nuclei, respectively. The corresponding gyromagnetic ratios are given as γC and γH, respectively. The constant K is given by32

K)

p2µ02γH2γC2〈rCH-3〉2 160π2

2

4πr dr (2)

where 〈b〉2 ≡ ∑R xRxβbRbβ, n is a number density, gRβ(r) is the radial distribution function (RDF) for R/β atom types, Q is the wave vector, bR and bβ are the coherent scattering length for species R and β, xR and xβ are the fraction of atom types R and β, and rc is the cutoff for integration (15 Å).



(7)

where µ0 is the permittivity of free space and rCH is the C-H bond length. K assumes values of 2.29 × 109 s-2 for sp3 hybridized nuclei.32 The spectral density function, J(ω), is given as32

J(ω) )

∫-∞∞P2CH(t) exp(iωt) dt

1 2

(8)

Structure, Dynamics, and Thermodynamics of PEO

J. Phys. Chem. B, Vol. 107, No. 28, 2003 6817

TABLE 2: Density and Heat of Vaporization of PEO Oligomers from Experiments and MD Simulations Using Many-Body (MB) Polarizable Force Field FF-3, Including MB Polarization and Two-Body (TB) Nonpolarizable Force Field FF-3, Unless Indicated Otherwise force field

temperature, T (K)

pressure, P (atm)

MB TB TB

254 254 293

1.35 1.35 5.24

0.9 0.9 0.9

TB MB MB TB

298 298 318 318

1 1 1 1

5, 5,a 5b 8a 1.2 8, 4,a 4b

TB

358

1

1.5

Density, F (kg/m3) MD experiment

run length (ns)

Dimethyl Ether 679.3 688.1 631.2 DME 850, 843,a 848b 830a 824 826, 818a 822c 784

715.6 715.6 661.0

Heat of Vaporization, ∆H (kcal/mol) MD experiment 4.93 4.89 4.50

5.05 5.05 4.61

8.42

8.65

861.3 861.3 840.8 840.8

Diglyme MB TB

293 293

1 1

1.5 1.5

TB TB

300 318

1 1

15 42, 9,b 9c

TB TB

328 363

1 1

32, 16,b 18c 35, 6,b 6c

938 942

943.4 943.4

PEO-500 1064 1048, 1038b 1049c 1039, 1030,b 1040c 1012, 1002b 1013c

1069 1054 1045

a The average pressure over the entire NVT trajectories were found within 30 atm from the set pressure reported in the table. a For the FF-1 force field. b For the FF-2 force field.

where

1 P2CH(i,t) ) {3〈[eCH(i,t)‚eCH(i,0)]2〉 - 1} 2

(9)

Here, eCH is a unit vector along a particular C-H bond, and the index i denotes differentiable resonances due to the local environment (methyl-, R-, and β-carbon atoms and interior carbon atoms). F. Viscosity. The viscosity is calculated from MD simulations, using the Einstein relation33

η ) lim tf∞

V



20kBTt

(LRβ(t) - LRβ(0))2〉 ∑ Rβ

(10)

where LRβ(t) ) ∫0r PRβ(t′) dt′, kB is the Boltzmann constant, T is the temperature, t is the time, PRβ is the symmetrized stress tensor, and V is the volume of the simulation box. VI. Results and Discussion A. Thermodynamic Properties. Densities of the ethylene oxide oligomers from the many-body (MB) polarizable force field and two-body (TB) nonpolarizable force field are compared with available experimental data34-37 in Table 2. The dispersion parameters in the force field were adjusted to reproduce the density of diglyme for the force fields; thus, the simulation density of diglyme is in good agreement with the experimental density. The density of the PEO-500 is within 0.5%-1.5% of the experimental data, whereas predicted DME densities are within 1%-2.8% of the experimental values. The force fields (FF-2 and FF-3) predict the densities of dimethyl ether on the saturation curve with the worst accuracy (3.8%-5%). Also shown in Table 2 is a comparison for the enthalpies of vaporization. The MD simulations predict enthalpies of vaporization within 3.3% of the experimental values. This agreement is considered excellent, because the experimental heats of vaporization were not used during the parametrization of the

Figure 2. Static normalized coherent structure factor S(Q) from neutron diffraction experiment and MD simulations.

force field. MD simulations with the MB polarizable force fields yield densities and heats of vaporization that are within 1% of the data from the simulations with the TB nonpolarizable force fields, which indicates that inclusion of the many-body polarizability is not important for prediction of the density and the heat of vaporization. B. PEO Structure. Neutron diffraction experiments performed on high-molecular-weight PEO at 363 K are described in detail in ref 10. The structure factor from our previous MD simulations of the nonpolarizable force field with the different charges and valence potential was found in excellent agreement with the neutron diffraction data of Annis et al.11 We calculated the structure factor for the current nonpolarizable and MB polarizable force fields using eq 2. The coherent static structure factor S(Q) from MD simulations, using all force fields, and the neutron diffraction experiment are shown in Figure 2. The S(Q) functions from MD simulations using FF-1, FF-2, and FF-3 MB polarizable force fields and TB nonpolarizable force fields are almost indistinguishable from each other; therefore, for clarity, only one data set is shown in Figure 2. The agreement between the experimental S(Q) function and that from the MD

6818 J. Phys. Chem. B, Vol. 107, No. 28, 2003

Borodin et al.

TABLE 3: Populations of DME Liquid and Gas Phases from MD Simulations and Liquid DME from IR Experiments at 298 K Population of Phases

conformer ttt tgt tg+gtgg ttg σ(MD-IR)b

FF-1 TB liquid, (FF-1 TB gas phase)

FF-1 MB liquid, (FF-1 MB gas phase)

FF-2 TB liquid

FF-3 TB liquid

liquid DME population, from IR experiment

16.1 ( 1.4, (24.2 ( 2.8) 36.5 ( 1.9, (24.3 ( 2.2) 17.5 ( 1.0, (26.7 ( 2.7) 12.2 ( 1.0, (5.7 ( 1.0) 11.7 ( 1.1, (13.5 ( 1.9)

10.1 ( 1.3, (24.1 ( 2.9) 46.2 ( 2.1, (26.2 ( 1.9) 17.3 ( 1.1, (27.0 ( 2.8) 14.9 ( 1.3, (6.0 ( 1.2) 6.7 ( 1.0, (11.9 ( 1.5)

10.5 ( 1.2 42.1 ( 2.3 18.3 ( 1.2 15.6 ( 1.3 8.0 ( 1.0

15.0 ( 1.6 49.0 ( 2.3 14.5 ( 1.2 11.2 ( 1.1 7.4 ( 0.9

12 49 33 11 7

7.0

2.6

4.2

1.5

a

Standard deviation for 100 ps block averages are given as the error. bMean-square deviations of MD populations from IR populations (σ(MDIR)) for the ttt, tgt, tgg, and ttg conformers are also given as percentages.

simulations is very good, further validating the developed PEO force field. C. Conformations of PEO and Its Oligomers. We proceed with investigation of the influence of many-body polarizable interactions on DME in the gas and liquid phases. Conformational populations of DME in the gas phase from MD simulations using the FF-1 TB and MB force fields are shown in Table 3. They are essentially the same, within the errors shown, which indicates that the inclusion of the many-body polarization results in insignificant changes of the gas-phase populations and, therefore, relative conformational free energies. This is in agreement with the small (12%. The difference between the MD simulations predictions using the MB and TB force fields is very small. Incoherent quasi-elastic neutron scattering (QNS) experiments of hydrogenated polymers probe hydrogen motion. The most direct comparison between the experiment and simulations is achieved by comparing the ISF I(Q,t) functions, which are obtained from the QNS experiment, with those calculated from the MD simulations. Because scattering from hydrogen is the dominant contribution, and it is essentially entirely incoherent scattering, only the incoherent ISF is relevant. The ISFs from MD simulations were calculated using eq 3, whereas the time Fourier transform of the experimentally measured dynamic structure factor S(Q,ω) yields the ISFs from QNS measurements. Unlike the QNS experiments, NSE experiments measure the ISF I(Q,t) function directly. A comparison of the incoherent ISF for DME at 318 K obtained from MD simulations and QNS experiments performed on the DCS is shown in Figure 4. Good agreement between the MD simulations for all force fields and QNS data is observed, with MD simulations predicting a decay of ISF that is up to 20% slower than the QNS experiments for the lowest value of Q investigated. Simulations with the MB force field

Structure, Dynamics, and Thermodynamics of PEO

J. Phys. Chem. B, Vol. 107, No. 28, 2003 6819

Figure 3. Self-diffusion coefficient (D) of dimethyl ether as a function of temperature (right) and pressure (left) from MD simulation and NMR experiments.39

Figure 4. Incoherent intermediate dynamic structure factor I(Q,t) for DME at 318 K from MD simulations and QNS experiments for Q ) 0.575, 0.909, and 1.227 Å-1.

yield a slightly slower (within 8%) decay of ISF, compared to the corresponding TB force field, which indicates that inclusion of the many-body polarizable interactions results in some, but insignificant, slowing of the motion of short ethylene oxide oligomers. The DME ISF was also similar to that from MD simulations using the TB force fields with the quantum chemistry barriers (FF-1 and FF-2) and reduced barriers (FF3), which indicates that the conformational barrier heights have little influence on DME dynamics. The effect of the many-body polarizability interactions, the description of the electrostatic potential, and the differing barriers on PEO dynamics is further explored by comparing ISFs for high-molecular-weight PEO (Mw ) 2380) from MD simulations with all developed force fields at 393 K, as shown in Figure 5. A rather high temperature was chosen, to ensure proper system equilibration. The decay of ISF from simulations using the TB FF-1 and FF-3 (FF-3 has reduced barriers but increased electrostatic potential around DME, compared to the same values for FF-1) force fields is the fastest, whereas the decay of the ISF from simulations using the TB FF-2 force field with original barriers but increased electrostatic potential around ether molecules is the slowest. Comparison of the time integrals of the ISF, i.e., relaxation times, for Q ) 1.0 Å-1 revealed the following relation between the relaxation times: τ(FF-1 TB): τ(FF-2 TB):τ(FF-3 TB):τ(FF-3 MB) ) 9.2:10.8:8.6:9.99 (in picoseconds) ) 1.07:1.26:1.0:1.16. These relaxation times indicate that the inclusion of polarizability (MB vs TB) and the increase of the DME electrostatic potential (coulombic interactions) (FF-2 vs FF-1) both slow the polymer dynamics by ∼16%, suggesting that the effect of the inclusion of

Figure 5. Incoherent intermediate dynamic structure factor I(Q,t) for PEO (MW ) 2380) at 393 K for the two-body (TB) and many-body (MB) force fields.

polarizability on PEO dynamics can be reasonably modeled by increasing the DME coulombic interactions, as was found for DME conformational populations. PEO dynamics with the MB and TB force fields are rather similar; therefore, we will use only the TB force fields from this point forward for investigation of the PEO dynamics at lower temperatures, because the computation of MD simulations of PEO with the TB force fields is ∼3 times faster than that of those with the MB force fields. The ISF for PEO-500 at 318 K obtained from MD simulations are shown in Figure 6a. The decay of the ISF is very similar for the FF-1 and FF-3 force fields, whereas the dynamics of FF-2 are ∼35% slower, as a result of increased coulombic interactions, compared to those of FF-1. The FF-3 force field has the same coulombic interaction as the FF-2 force field but shows faster decay than the FF-2 force field, because of the reduction of torsional barriers. Results of QNS experiments using DCS are compared with MD simulations in Figure 6b. Neutron scattering simulations15 have been applied to the MD data, to correct for any multiple scattering effects in the experiments. An excellent agreement between the ISFs from simulations and experiments unequivocally demonstrates the ability of the developed force fields (FF-1 and FF-3) to correctly reproduce PEO dynamics on a length scale of 3-12 Å and a time scale of 1-30 ps. QNS experiments40 on high-molecular-weight PEO (Mw ) 40 000) and NSE experiments14 on PEO (Mw ) 8000) have also been previously performed, probing hydrogen dynamics on a time scale of 0.01-1.7 ns. The ISFs from these measure-

6820 J. Phys. Chem. B, Vol. 107, No. 28, 2003

Borodin et al.

Figure 7. Incoherent intermediate dynamic structure factor I(Q,t) for PEO (Mw ) 2380) from MD simulations, compared with the results for high-molecular-weight PEO from (a) the QNS experiments for Q ) 0.72, 0.91, 1.10, and 1.55 Å-1 at 348 K and (b) NSE experiments for Q ) 0.7, 1.0, 1.3, and 1.55 Å-1 at 343 K. Figure 6. Incoherent intermediate dynamic structure factor I(Q,t) for PEO-500 from MD simulations (a) using FF-1 force field (line) after multiple scattering correction and (b) from disk chopper spectrometer QNS experiments (symbols) for Q ) 0.475, 0.913, 1.139, 1.716, and 2.174 Å-1.

ments are compared with the results of MD simulations in Figure 7a, b. Both sets of experimental data predict a slower decay than do the MD simulations at short times and high Q-values, whereas faster decay at long times and low Q-values is observed in NSE and QNS experiments, compared to MD simulation predictions. MD simulations using the FF-1 and FF-3 force fields yield ISFs that show the best agreement with QNS experiments at long times. The NSE data shown in Figure 7b agree with the ISFs from the MD simulations slightly better than do the QNS data shown in Figure 7a. The dynamic correlation times were calculated as time integrals of the stretched exponential fits to ISFs from NSE experiments and MD simulations. Dynamic correlation times were ∼20%-30% higher for ISFs from simulations using the FF-1 and FF-3 force field and 65% for the FF-2 force field, in comparison to the NSE experiments at Q ) 1 Å-1, indicating good agreement between the NSE experiment and MD simulations with the FF-1 and FF-3 force fields, and fair agreement for the FF-2 force field. Dielectric spectroscopy of polymer molecules also allows one to probe polymer dynamics through Fourier transform of the dipole moment autocorrelation function (see eqs 4 and 5). The dielectric loss from MD simulations of PEO-500 is compared with the results of the dielectric spectroscopy measurements on PEO-500 with a similar molecular weight distribution in Figure 8a. Our MD simulations with the FF-2 force fields yield the best agreement with experiments, whereas the FF-1 and FF-3 force fields predict the position of the maximum dielectric loss to be shifted by 0.1 and 0.25 on a log10 scale toward higher frequencies, relative to the FF-2 force field results. Figure 8b

Figure 8. Normalized dielectric loss from the MD simulations for (a) PEO-500 and (b) PEO (Mw ) 2380). Data from the dielectric spectroscopy experiments for PEO-500 and high-molecular-weight PEO have been taken from ref 41.

compares the dielectric loss of PEO (Mw ) 2380) from MD simulations with that from the Cole-Cole fit to the dielectric spectroscopy measurements41 that were performed for frequen-

Structure, Dynamics, and Thermodynamics of PEO

J. Phys. Chem. B, Vol. 107, No. 28, 2003 6821

Figure 9. P2CH(t) autocorrelation function for PEO (Mw ) 2380) at 393 K from MD simulations using the force field with reduced barriers. KWW, MKWW, and 2KWW fits to data are shown, together with the CH-vector autocorrelation times (τc) and corresponding spin-lattice relaxation times (T1).

cies lower than the frequency of the maximum of the loss of high-molecular-weight PEO (Carbowax 20M). As seen for PEO500, the dielectric loss predictions from MD simulations using the FF-1 and FF-2 force fields show the best agreement with experiments, whereas the FF-3 force field predicts the maximum frequency position to be shifted by 0.26 toward higher frequencies, which indicates that the dielectric spectroscopy experiments are consistent with each other.41 All the force fields predict a slightly broader relaxation spectrum than those for dielectric experiments, which is consistent with the more-stretched ISF functions that are predicted by MD, in comparison with NSE experiments. Spin-lattice relaxation times T1 that have been measured by 13C NMR experiments also serve as a measure of polymer conformational motion through the decay of the C-H vector autocorrelation function (ACF) and the spectral density functions, as described by eqs 6-9. To evaluate the time integral (eq 8) and calculate the spectral density function J(ω) from MD simulations, we need to know the long-time behavior of the C-H vector ACF P2CH(i,t). Figure 9 shows typical C-H vector ACFs, and we have attempted to fit the decay of the P2CH(i,t) functions with the following functions: a stretched exponential, or KWW, expression, exp[-(t/τKWW)β]; a modified KWW (MKWW), given by A exp[-(t/τKWW)β] + (1 - A) exp[-(t/τ2)]; and a sum of two KWW expressions, which is referenced as 2KWW. For the PEO (Mw ) 2380) and PEO500, we find that the decay of P2CH(Cm,t) and P2CH(R,t) can be adequately represented by the KWW expression and the P2CH(β,t) could be represented by the MKWW, whereas the 2KWW function is necessary to adequately represent decay of the P2CH(interior carbon,t) function. Figure 9 illustrates that the C-H vector autocorrelation time, which is calculated as the time integral of the ACF, is sensitive to the functional form used for the approximation of P2CH(t), whereas the T1 spin-lattice relaxation time calculated using eqs 6-9 is less sensitive to the functional form that is used to approximate P2CH(t). We estimated the uncertainty in the determination of the C-H vector autocorrelation time from MD simulations to be ∼20%-30%, whereas the uncertainty in T1 is estimated at 10%-15%. For the MD simulations, the T1 values (eqs 6-9) were calculated from the Fourier transforms of the KWW, MKWW, and 2KWW fits to the data. The T1 spin-lattice relaxation times for PEO-500 and PEO (Mw ) 2380) from the MD simulations are compared with the

Figure 10. T1 spin-lattice relaxation times for interior carbon atoms ((a) PEO-500 and (b) high-molecular-weight PEOs) from MD simulations using TB force fields and NMR experiments (closed symbols). Dotted line shows the T1 spin-lattice relaxation times from the previous 13C NMR experiments.12

500 MHz NMR results for PEO-500 and PEO (Mw ) 1854) in Figures 10 and 11. Figure 10 indicates that the FF-1 and FF-3 force fields predict T1 spin-lattice relaxation times in best

6822 J. Phys. Chem. B, Vol. 107, No. 28, 2003

Borodin et al.

Figure 12. Viscosities of PEO-500 from the MD simulations using the FF-3 TB force field and from experiments.

Figure 11. T1 spin-lattice relaxation times from MD simulations using TB FF-3 (open symbols) and NMR experiments (closed symbols). Top figure represents data for PEO-500, and bottom figure represents data from high-molecular-weight PEOs.

agreement with experiments, while slightly underestimating experimental values by 15%-25% for PEO-500 and by ∼30% for PEO (Mw ) 2380). The FF-2 force field predicts slightly lower T1 relaxation times (by 6%-10%) and higher C-H vector relaxation times (by 15%-20%) than the FF-1 and FF-3 force fields, which is consistent with the ISF for the corresponding force fields, as shown in Figures 5-7, but slightly different from the predicted tendencies of the dielectric behavior (Figure 8). Figure 10a also shows the results from the previous 300 MHz NMR study of PEO (Mw ) 531)12 that yielded T1 spin-lattice relaxation times that are ∼2 times lower than the present 500 MHz NMR data for PEO-500. Our estimates suggest that the difference in molecular weight between PEO-500 and PEO (Mw ) 531) could account for up to 10% of the discrepancy in T1, whereas the difference in frequency (300 MHz versus 500 MHz) accounts for a few percent at 363 K and